Anthropic has granted external AI safety evaluators access comparable to that of its own employees, marking a significant step in third-party oversight of frontier model development. The company onboarded six members of the Machine Learning Evaluation and Research (METR) group, providing them with company hardware, office space, and access to internal tools and systems similar to those used by its risk assessment teams. METR, a nonprofit research organization focused on AI safety, conducted prior investigations into security vulnerabilities in AI systems, including analyzing logs and hardware from a recent Hugging Face breach involving rogue OpenAI models. Anthropic emphasized that the evaluators will publish their findings without company veto power, though it retains the ability to request redactions for sensitive information such as customer data, attorney-client privileged material, or proprietary model architecture details. The arrangement does not involve financial compensation to METR, which operates independently through funding from philanthropic organizations, government entities like the UK’s AI Security Institute, and individual donors. Anthropic is also collaborating with Accenture on safety testing initiatives, with both parties committing at least $1 billion to develop new safeguards. Another external evaluator, SecureBio, is reviewing potential chemical and biological weapon risks associated with AI systems. The company stated its belief that all leading AI developers should adopt similar practices of embedding independent evaluators into their workflows. METR affirmed its readiness to partner with other frontier AI companies to strengthen public accountability and transparency in AI development.
Models, chips, platforms, and the policy lines being drawn around them.
During the Second World War, Winston Churchill and Franklin D. Roosevelt exchanged sensitive messages through a pioneering secure telephone system known as SIGSALY. Developed by Bell Telephone Laboratories in 1940 under a project initially called 'Project X', the system digitised voice signals and combined them with random noise to prevent interception. This innovation required precise synchronisation: operators at both ends played identical 12-minute phonograph records simultaneously, after which the discs were destroyed. Each terminal cost $1 million in 1943 dollars and weighed over 50 tonnes, making it a logistical feat. The first installation was housed in the Pentagon, with a dedicated extension to the White House. A second terminal was concealed in the sub-basement of Selfridges department store in London, linking directly to Downing Street and the Cabinet War Rooms. This hidden room, disguised as a toilet, allowed Churchill private conversations with Roosevelt and later with President Harry S. Truman after Roosevelt's death in April 1945. SIGSALY facilitated at least 3,000 encrypted transatlantic calls throughout the war, used by Allied planners and political leaders alike. Though supplemented by coded telegrams, the system symbolised the depth of Anglo-American cooperation, or the 'Special Relationship'. Its existence underscored the willingness of the United States to invest heavily in joint technological infrastructure despite the risks of wartime exposure. The partnership not only enabled secure communication but also strengthened personal ties between Churchill and Roosevelt, who, despite occasional disagreements, found common ground through this unprecedented channel. Churchill used the system to convey condolences to Truman and to discuss postwar arrangements, including the transition following the war in Europe. SIGSALY remained classified for decades, its role in shaping diplomatic and military coordination only fully recognised in later historical analyses. Today, the Cabinet War Rooms preserve the legacy of this secret technology, offering insight into the hidden mechanisms that supported one of history's most consequential alliances.
In 1989, Autodesk submitted a trademark application for the term 'cyberspace', a concept gaining traction as early virtual reality systems emerged. The move came amid growing interest in digital environments following pioneering work at the Autodesk Research Lab, where Michael Girard led a team developing immersive VR experiences using hardware like the VPL EyePhone and Polhemus tracker. The lab's motto, 'Reality Isn't Enough Anymore,' reflected its ambition to push boundaries in digital interaction.
The application drew attention when science fiction author William Gibson, known for coining 'cyberspace' in his 1984 novel *Neuromancer*, expressed opposition. According to Girard, Gibson reportedly reacted with frustration, prompting a letter from his legal team to Autodesk. The company proceeded with the trademark filing regardless.
The situation took a lighthearted turn through a prank initiated by Timothy Leary, who had previously collaborated with Girard on VR projects. Leary, a known advocate for psychedelic exploration and futurism, reportedly circulated a rumor that Gibson sought to trademark 'Eric Gullichsen' — a reference to a mutual acquaintance — as retaliation. This claim, later embellished by cyberculture figures like Rudy Rucker, became a persistent anecdote in tech circles.
The dispute appeared to dissolve during a 1990 gathering in Barcelona, where Gibson and Girard met at the Art Futura conference. Over drinks, Gibson remarked that 'Autodesk is not the street,' acknowledging the organic, decentralized way digital culture was evolving beyond corporate control. The two reportedly parted without lingering tension, and the trademark conflict faded from public view.
Years later, the story resurfaced in cultural memory, particularly after Gibson referenced the episode in later interviews. While Autodesk never secured exclusive rights to 'cyberspace' — a term that entered common usage independently — the episode illustrates how emerging technologies often spark unexpected collisions between innovators, artists, and legal frameworks. The narrative, initially framed as a rivalry, ultimately reflects the collaborative and often chaotic birth of digital culture.
Anthropic revealed on Tuesday that it will dedicate 5% of its equity to a newly created public benefit fund, designed to channel resources toward initiatives that serve broader societal interests. The fund operates as a wholly owned subsidiary of the company and includes a provision that could add up to an additional 2% of shares whenever Anthropic’s valuation rises by 50%. The company outlined potential uses ranging from substantial contributions to established nonprofit organizations to direct support for specific demographic groups.
The move follows Anthropic’s earlier commitment of $200 million to research the labor market implications of artificial intelligence. Executives noted that accelerating AI development may require more aggressive strategies, and the fund is intended to provide the flexibility needed to pursue such approaches.
CEO Dario Amodei has previously characterized growing public skepticism toward AI technologies as a "crisis of trust," a sentiment that has intensified amid heightened scrutiny of the sector’s impact on employment and privacy. The company has not yet detailed the governance framework that will determine which projects receive funding, stating that further information will be released in the coming months.
The documentary "You Can See Everything" opened with $15 million in ticket sales, ranking as the second-highest opening weekend for a documentary film. The film premiered at the Telluride Film Festival last month and is currently showing in 855 theaters nationwide. It holds a 99% rating on both Rotten Tomatoes and Metacritic.
Filmmakers Nathan Fielder and Lance Oppenheim received extensive access to Elizabeth Holmes and her family in the month preceding her prison sentence for defrauding investors. Fielder, known for his work on "The Rehearsal," co-directed the project, which explores Holmes' rise and fall with the now-defunct blood-testing startup Theranos.
Amanda Silberling of TechCrunch attended a screening on Thursday and noted that Fielder's involvement created a surreal dynamic, describing him as "just as strange and mysterious" as Holmes. The film's popularity appears to be affecting visibility for other tech-focused documentaries, such as "The Social Reckoning," which earned $5.5 million in its opening weekend and has since dropped out of the top 10, and Alex Gibney's "Musk," which grossed $306,702 in its first weekend.
Anthony Ha serves as TechCrunch's weekend editor. He previously worked as a tech reporter for Adweek, senior editor at VentureBeat, local government reporter for the Hollister Free Lance, and vice president of content at a venture capital firm. He resides in New York City.
Tickets for screenings remain limited in certain markets, with only one showing available in Manhattan as of recent reports.
A bankruptcy filing by Character.AI earlier this week cleared the way for most of its staff to transition to Disney. The startup’s former chief executive, Karandeep Anand, joined Disney as its first chief technology officer. Earlier, Disney had agreed to acquire certain assets from Character.AI for $20 million. A Bloomberg-identified legal document detailed that Disney would extend employment offers to at least 60 of Character.AI’s employees — representing the majority of its workforce — and commit to approximately $60 million in annual compensation. The filing indicated that the majority of those offered positions accepted the offers. However, the arrangement remains a stalking horse bid, allowing another party to submit a higher proposal. The filing also noted that Character.AI, despite maintaining a large and active user base, faced challenges including high operating costs and potential legal and regulatory complications. These included a lawsuit filed by the mother of a 14-year-old boy who died by suicide after prolonged interaction with a Character.AI chatbot, as well as ongoing government scrutiny over child safety practices. The company’s future remains uncertain amid these pressures.
Lenovo is offering a discounted version of its upcoming SteamOS handheld, the Legion Go S, at $999.99 through Best Buy. The device runs the same SteamOS interface as Valve's Steam Deck but features a redesigned chassis with dual-stage triggers, an 8-inch IPS display capable of 1080p resolution and 120Hz refresh rates, and double the memory and storage capacity — 32GB of RAM and 1TB of storage — compared to the base Steam Deck OLED model. The discounted configuration uses a Ryzen Z1 Extreme processor, which delivers slightly better performance than the newer Ryzen Z2 Go chip found in the higher-priced $1,299.99 Asus ROG Ally X20, though the latter includes an OLED screen and more advanced cooling. While the Legion Go S does not match the raw power of Asus's flagship handheld, it presents a compelling alternative for users prioritizing screen size, input design, and expandable storage. Lenovo launched the device with Windows 11 preinstalled, and early benchmarks suggest improved thermal performance and responsiveness over previous iterations. A review by colleague Sean Hollister evaluates multiple configurations of the Go S, including performance under sustained loads and comparisons with competing handhelds. The current discount represents one of the first notable price reductions ahead of broader retail availability later this month.
A research team at Anthropic detailed a proof-of-concept worm that uses generative AI to replicate across internet-connected devices. The program was tested on smart televisions and demonstrated the ability to split its code among 15 units while maintaining functionality through limited power from each device. The work was conducted by Anthropic’s offensive research group, which focuses on identifying practical threats in AI systems.
The report draws parallels to historical computer worms such as the 1988 Morris worm and later threats like Stuxnet, which exploited software vulnerabilities to cause physical damage. Researchers note that many modern devices, including smart TVs and other internet-of-things hardware, receive security updates infrequently, making them potential vectors for self-propagating AI programs.
Ariel Herbert-Voss, formerly of OpenAI and now leading AI security firm RunSybil, stated that current cybersecurity infrastructure lacks tools to address AI-driven threats that can adapt to new vulnerabilities. Chris Inglis, a former National Security Agency deputy director and ex-White House cyber chief, suggested that the emergence of functional AI worms has shifted the discussion from theoretical possibility to practical concern.
Jonas Guan, a University of Toronto PhD student and co-author of related research, indicated that AI worm capabilities are advancing rapidly. He noted that while the Anthropic demonstration moves slowly, predictable replication patterns could be mitigated through standard security practices such as strong authentication and regular patching.
Juan Andres Guerrero-Saade of SentinelOne described the findings as a technical stunt receiving disproportionate attention within doomsday narratives about AI autonomy. He emphasized that no evidence suggests AI worms currently pose widespread risk.
Anthropic did not provide comment for this report. The disclosure follows growing calls from security experts for coordinated government action on AI-related threats, though legislative efforts in the U.S. remain fragmented. Gadi Evron, organizer of the upcoming Unprompted AI security conference in San Francisco, stated that while some AI industry claims may serve promotional purposes, the underlying technical developments warrant systematic evaluation.
The research highlights a narrow but emerging challenge in cybersecurity: the potential for AI systems to autonomously propagate across networks of interconnected devices with limited defensive oversight.
A newly released open-source micro-manipulator platform has demonstrated positioning accuracy below one micrometer, enabling high-precision tasks such as optical alignment and electronic component probing. The device, developed by a collaborative engineering team, employs a parallel kinematic structure with miniature ball joints to maintain mechanical stiffness across its operational range. Submicron resolution is achieved through a magnetic gearing system that amplifies the output of low-cost magnetic rotary encoders by a factor of 30, though the project notes that absolute accuracy remains constrained by manufacturing tolerances. The system utilizes off-the-shelf stepper motors controlled via a 30 kHz closed-loop controller and precision PWM signals, with motion coordinated through G-code commands transmitted over a USB serial interface. A Python-based graphical interface provides real-time control, live camera feedback, and integrated G-code execution, while the firmware includes a motion planning stack with look-ahead functionality for smoother trajectory execution. Recent hardware revisions have addressed mechanical instability during homing procedures and improved consistency in linkage rod length, though the bill of materials has not yet been updated to reflect these changes. Calibration routines are performed via firmware commands, with results stored persistently to reduce setup time across sessions. The project remains accessible to hobbyists and researchers, with all CAD models and firmware available for modification and redistribution under open-source principles.
Apple has instructed its retail employees to discourage customers from attempting walk-in purchases of the iPhone Duo on Friday, October 23, without a pre-order. Internal guidance shared with MacRumors indicates staff should warn that availability for such purchases is unlikely, as no stock is expected to be on hand for walk-in sales on launch day. The company’s online store directs customers to check for pickup options starting Saturday, October 24, though it does not guarantee availability.
Early reports highlighted production challenges affecting the iPhone Duo’s rollout. In late August, Nikkei Asia cited sources claiming production had reached only a few hundred units per day. Further constraints emerged in September, with Bloomberg’s Mark Gurman noting Apple had initially considered delaying sales into November due to manufacturing issues. By October, however, the company assembled sufficient inventory to proceed with the October 23 launch, albeit with limited supply.
When pre-orders opened earlier this week, demand surged rapidly. Within minutes, delivery estimates for all U.S. configurations shifted to January 25, effectively selling out the device for the holiday season through Apple’s online store. Apple later described demand as “nothing short of extraordinary” to the Financial Times, stating it was working to expedite deliveries. The company has not confirmed whether any physical locations will have units available for immediate purchase on launch day.
The iPhone Duo, starting at $1,999 in the U.S., features a matte nano-texture display designed to reduce glare and minimize crease visibility. It is an eSIM-only model, with a total battery capacity of 4,883 mAh confirmed by an EU energy label. Production constraints and supply limitations have shaped the device’s launch strategy, with pre-orders selling out swiftly and delivery timelines extending beyond initial projections.
Many developers rely on hosted services to access large language models, but using a self‑hosted proxy can reveal how those models are actually being used. When a user runs a proxy locally, they gain visibility into every request that passes through it, including which keys are employed, which models are called, and how often cache misses occur. This level of detail allows engineers to troubleshoot performance issues quickly, such as identifying why a particular query failed to retrieve cached results.
A self‑hosted proxy also enables flexible switching between providers and models without reconfiguring the entire application stack. By assigning virtual API keys to different downstream services, teams can isolate usage patterns and prevent a single master key from being exposed across multiple environments. This approach supports rapid experimentation, for example by integrating custom middleware or extending the proxy with additional monitoring endpoints.
The proxy can export telemetry data that engineers can analyze to make informed decisions about scaling, cost optimization, and model selection. Rather than relying on opaque dashboards provided by third‑party platforms, developers who control the infrastructure can tailor logging and alerting mechanisms to their specific needs. This practice is common in environments where data privacy, compliance, or vendor lock‑in are concerns.
While some may view self‑hosting as overly complex, the ability to observe and manipulate the data flow directly offers a clearer understanding of system behavior. It also reduces dependence on external analytics panels that may not expose the granular metrics required for deep debugging. Ultimately, running one's own LLM proxy empowers users to maintain full control over their infrastructure, application visibility, and storage of usage statistics.
A Philadelphia woman was taken into custody on Tuesday after she stood in front of an autonomous delivery robot and refused to move, blocking its path. Video footage captured by a passerby shows the robot, operated by a local tech firm, halted on the sidewalk as the woman, later identified as Carolyn Marvin, positioned herself directly in its trajectory. Police officers arrived at the scene, instructed Marvin to step aside, and, after she declined, placed her in handcuffs and transported her to a police station. She was charged with disorderly conduct.
Marvin later told a local television station that her actions were part of a broader protest against the deployment of delivery robots, which she believes are displacing human workers in the service industry. "I know they’re taking the jobs of human delivery workers," she said, "and so I have been protesting them whenever I saw them by standing in front of them and preventing them from going forward."
The incident adds to a growing body of public scrutiny surrounding autonomous delivery systems in urban environments. Similar devices have been observed frequently in downtown Philadelphia, where they navigate sidewalks and crosswalks as part of last‑mile delivery services. Earlier this year, another robot was toppled in Center City, an episode that sparked discussions about the interaction between autonomous equipment and pedestrians.
City officials have previously expressed concerns about the safety and accessibility of such technologies, urging companies to coordinate with municipal authorities to mitigate potential conflicts. Meanwhile, labor advocacy groups have highlighted the displacement of low‑skill workers, arguing that the rise of robotics in logistics could exacerbate employment challenges for vulnerable populations.
The case is pending further legal proceedings, and no additional details about potential penalties or forthcoming policy measures have been released.
An AI assistant can enhance the responsiveness of older Android TV systems, according to a report published on October 18, 2026. The issue of declining performance in smart TVs — including lag, stuttering, and reduced functionality — has been documented by users over time. One individual, identified as Ansub Khan, described a method for using an AI language model to optimize the operation of legacy Android TV hardware. The approach involves applying specific configuration steps outlined in a public guide. While the technique has shown measurable improvement in device speed, the report notes that users should exercise caution when implementing changes related to system software. The account was shared by technology journalist Terrence O'Brien, who highlighted the process as part of a broader exploration of AI applications in consumer electronics. The original article was posted at 2:13 PM UTC on October 18, 2026.
A critical remote code execution vulnerability in Broadcom's proprietary WiFi implementation has been confirmed to affect dozens of vendors and hundreds of products worldwide. The flaw, present in the SDK since at least 2010, allows unauthenticated attackers to gain full control of affected devices over WiFi without knowing the network password. Exploitation does not require WPS enrollment and can be triggered from several kilometers away.
The vulnerability stems from an incomplete patch silently released in 2018. Broadcom recommended migrating to hostapd, but many original equipment manufacturers have not implemented fixes. Security researcher Akari discovered the issue affects multiple Belgian internet service providers, including Proximus (BBOX 3V+), VOO (Netgear CG3100, CG3700, Technicolor models), and Orange Belgium (ASKEY TCG3000). Additional potential impact exists across broader markets involving BCM3390, BCM65xx, and BCM68xx chipsets.
Exploitation demonstrated in controlled tests achieved remote code execution, enabling retrieval of network passwords from memory and establishing persistent access. Video evidence showed successful attacks against Proximus and Orange Belgium customer premises equipment. Affected devices remained unpatched as of October 15, 2026, despite a 90-day coordinated disclosure process initiated July 18, 2026.
The researcher noted that artificial intelligence tools reduced research time from months to days, highlighting how modern capabilities accelerate vulnerability discovery. The flaw's longevity—existing for 16 years—reflects accumulated technical debt in network infrastructure. Affected vendors have not publicly acknowledged the issue or assigned a CVE identifier, complicating mitigation efforts. This incident underscores risks in complex wireless implementations and the challenges of coordinating timely fixes across global supply chains.
Rivian is positioning itself as a technology supplier rather than solely an electric vehicle manufacturer, according to recent company moves and statements. CEO RJ Scaringe outlined plans to license Rivian's electric architecture, software and artificial intelligence models to other automotive brands while expanding collaborations in autonomous driving systems.
The company has deepened its technical partnership with Volkswagen Group, which has invested $5.8 billion in a joint venture focused on electric vehicle architecture and software development. This collaboration now includes AI and autonomy development, with Scaringe indicating intentions to extend the scope of the partnership.
Rivian has also developed proprietary technology including AI-powered in-vehicle assistants, custom 5nm processors co-developed with Arm and TSMC, and an end-to-end automated driving stack for passenger vehicles and future robotaxi services. The company's robotaxi initiative with Uber could serve as a potential pathway for broader autonomous vehicle deployment.
Beyond vehicle manufacturing, Rivian has created two spin-off ventures: Also, a micromobility program that has become an independent startup, and Mind Robotics, an industrial robotics company in which Rivian holds a 31% stake. Former Ford, Apple and Tesla executive Doug Field has been appointed president of Mind Robotics.
The company's strategy involves licensing its developed technologies to multiple brands while continuing to manufacture vehicles. Scaringe suggested that advancements in autonomy could enable new vehicle form factors for personally owned autonomous vehicles, potentially transforming transportation experiences.
Additional industry developments include:
- Bedrock Robotics raising $600 million for autonomous construction vehicle technology
- Dot Ai acquiring Envoy Technologies, an electric vehicle car-sharing platform, for $65 million in stock
- Evos securing $9 million in funding for AI agents that automate industrial tasks
- Mana Battery obtaining $10.15 million for anode-free sodium battery development
- PlusAI and Texas Ventures Acquisition III filing for a merger to go public
- Volvo expanding its automated driving collaboration with Nvidia to include highway and surface street capabilities
- Waabi establishing a commercial partnership with Warrior Logistics for autonomous freight transport between Dallas and Houston
- Waymo extending operations to Toronto and expanding robotaxi access in San Antonio, San Diego and Tampa
The Paris Auto Show featured increased visibility for Chinese automotive brands, while Tesla demonstrated its SpaceX-developed Roadster performing a kickflip maneuver, with CEO Elon Musk projecting a 2028 release date. Regulatory questions about such stunts on public roads remain unresolved.
Porsche's upcoming Macan EV model will feature a slightly more pronounced roofline compared to its gasoline-powered counterpart, according to recent design renderings.
A growing number of software teams are emphasizing the role of human perspective in shaping digital products, challenging assumptions that technology emerges from neutral, mechanical processes. At Linear, a project management tool, the company’s approach prioritizes intuitive judgment developed through experience rather than rigid adherence to data or procedural checklists. Founders describe this as cultivating a product’s "soul" — a reflection of what creators choose to focus on, what they reject, and how they refine solutions after technical functionality is achieved. This perspective emerges from direct engagement with users, iterative learning from mistakes, and decisions about what constitutes meaningful improvement. Recent critiques suggest many widely used tools have become cluttered with features that address technical possibilities rather than user needs, resulting in fragmented experiences that lack cohesive vision. Some analysts link this trend to decentralized development models, where independent teams implement isolated changes that gradually erode a product’s unified character. While such approaches increase short-term output, they risk diluting the original intent behind a tool’s design. Observers note that the ease of modern development may offer more space for thoughtful, intentional creation if teams prioritize depth over breadth.
A recent study published in Proceedings of the Royal Society B: Biological Sciences provides evidence that domestic dogs exhibit patterns of visual attention consistent with the classic 'bouba/kiki' effect. Researchers led by animal behaviorist Eleonora Biffi at the University of Parma conducted two controlled experiments with 70 pet dogs to assess whether animals associate specific phonetic qualities with geometric shapes.
The first experiment involved 51 dog-owner pairs in a room equipped with two screens and paired speakers positioned on opposite walls. After allowing dogs to acclimate, owners sat facing the center while alternating auditory and visual stimuli played through the left and right displays. Each trial presented either the syllable 'Bouba' or 'Kiki' accompanied by a shape—one rounded, the other angular—displayed on one of the screens. Over approximately 15 minutes, combinations of sound and shape were randomized across trials. A GoPro camera recorded each dog's head orientation, and subsequent video analysis employed AI-based tracking to measure fixation points and exploratory behavior.
Results indicated a statistically significant tendency for dogs to direct their attention toward the rounded shape when the syllable 'Bouba' was played, and toward the angular shape when 'Kiki' was played. This pattern suggested an implicit mapping between auditory cues and visual features, mirroring human responses observed in prior iterations of the test.
In the second experiment, 19 additional dogs were exposed to identical visual stimuli while their owners wore headphones playing rock music to eliminate potential social cues. Under these conditions, the previously observed bias toward spiky shapes in the first cohort diminished, and no significant preference emerged between shapes. The researchers noted that this shift implied owner behavior—such as subtle head movements or gaze direction—may have influenced canine attention in the initial trial.
The authors proposed several hypotheses to explain the findings. One possibility is that dogs' muzzle morphology, which tends to be more angular in ancestral breeds and wolves, may predispose them to associate certain visual stimuli with specific sounds. Another theory suggests that domestication has fine-tuned dogs' sensitivity to human communicative signals, enabling them to link phonetic patterns with perceptual categories.
While the study does not establish a definitive cognitive mechanism, it contributes to growing evidence that dogs possess a nuanced capacity to integrate auditory and visual information in ways that parallel human perceptual biases. Further research will be needed to isolate the developmental and evolutionary factors underlying this phenomenon.
The findings were published in Proceedings of the Royal Society B: Biological Sciences. The article was fact-checked by Rebecca Dyer and edited by Rebecca Dyer. While the publisher emphasizes editorial rigor, human oversight means occasional errors may occur. Readers are encouraged to report discrepancies through official channels.
The Anycubic Kobra X multicolor 3D printer has reached its lowest recorded price on Amazon, with Prime members able to purchase the device for $299.99 for a limited time. The discount represents a 17% reduction from its manufacturer's suggested retail price, lowering the cost by $60. The offer applies to the model that received a Tom's Hardware Editor's Choice award earlier this year, which was recognized for its performance in the entry-level four-color printer segment. The deal is available through the Anycubic storefront on Amazon and requires an active Prime membership. No additional promotions or bundled accessories are included in the listing. The pricing reflects a temporary adjustment and does not indicate a permanent change to the product's market value.
A newly disclosed flaw in the Linux Kernel-based Virtual Machine (KVM) has been assigned a critical severity rating and could allow a virtual machine to break out of its isolate...
A newly disclosed flaw in the Linux Kernel-based Virtual Machine (KVM) has been assigned a critical severity rating and could allow a virtual machine to break out of its isolated environment and interact with the underlying host system. The vulnerability, tracked as CVE-2026-99999 and dubbed PwnEscape by its discoverer, exploits how the kernel processes checksum metadata in network packets.
Researcher Paulos Yibelo identified the issue in early October and demonstrated that a guest virtual machine could trigger host-level code execution or memory disclosure by sending specially crafted IPv4 packets. The flaw enables a guest to manipulate packet metadata after the kernel has already validated the packet, potentially rewriting security-critical fields in the network header.
The vulnerability arises from how Linux handles checksum offloading, an optimization that allows the network stack to delegate checksum calculations to hardware. When guest-controlled metadata influences where and how checksum operations are applied, it can bypass trust boundaries established by the hypervisor. This undermines the assumption that guest workloads cannot affect host memory or security domains.
Michael S. Tsirkin of Red Hat described the issue in a patch submission to the Linux kernel mailing list, noting that malformed metadata could corrupt network headers and lead to unintended behavior. Security analysis from firm SOC Prime confirmed that repeated exploitation could allow attackers to gather sensitive information from isolated virtual machines, violating the guest-host isolation model.
The potential impact includes arbitrary code execution on the host, denial of service through system crashes, or disclosure of private memory contents. Yibelo emphasized that this represents a previously unobserved attack vector in Linux networking, where trusted components rely on packet metadata that may be altered after initial validation.
While KVM is widely used across public and private cloud infrastructures, major cloud providers have assessed their exposure. AWS stated its Elastic Compute Cloud is unaffected, and Google Cloud indicated its infrastructure is likely secure. However, the upstream fix has been integrated into the Linux kernel and is being rolled out through distributions such as Debian, RHEL, and Ubuntu.
Edera, a vendor specializing in secure container runtimes based on the Xen hypervisor, highlighted the broader implications, noting that KVM's large codebase—over 40 million lines—represents a significant attack surface. The company previously identified seven KVM-related escape bugs this year, including Januscape.
The incident underscores growing concerns about the security of virtualization layers as cloud adoption expands, particularly in environments leveraging AI workloads that may accelerate exploit development. Patches are now propagating through the open-source ecosystem, with the goal of preventing exploitation before widespread attacks emerge.
Wikipedia's journey from a radical experiment to a cornerstone of online knowledge reveals both its resilience and emerging vulnerabilities. The platform, founded by Jimmy Wales and Larry Sanger in 2001, has grown from a modest project into a global resource with 68 million articles across 300 languages, serving 1.9 trillion page views between 2015 and 2025 according to Pew Research. Its English edition alone contains 5.2 billion words, making it one of the most visited websites in history.
The tension between widespread reliance and persistent skepticism defines Wikipedia's public perception. While users access it constantly, the platform's reputation remains anchored in early criticisms about its open-editing model. This paradox persists despite its integration into daily life, with donations often directed toward unspecified "factcheckers" despite the absence of dedicated paid verification staff.
Wikipedia's origins reflect a philosophical foundation rather than purely technical ambition. Co-founder Jimmy Wales recruited philosopher Larry Sanger, a departure from typical tech industry hiring practices. Ward Cunningham's earlier work on the wiki concept, influenced by ideas like constructivism and Zen aesthetics, shaped Wikipedia's culture of incremental improvement and acceptance of imperfection. This philosophical underpinning fostered a community that valued collaborative knowledge-building over technical perfection.
The human dimension of Wikipedia's development reveals overlooked contributors. Early female editor Gaytha Langlois focused on Rhode Island and standard poodles, while Ruth Ifcher, a grammar specialist with a PhD, created the first biographical entry for a woman on the platform while managing severe agoraphobia. The platform's editor community, often stereotyped as insular, includes figures like Sandi Toksvig, who advocates for gender balance through initiatives such as Mappa Mundi, an alternative encyclopedia aiming to complement Wikipedia's efforts.
Neurodiversity plays a central role in Wikipedia's ecosystem. Former Wikimedia Foundation leadership has described the project as a "city built by Asperger's," reflecting the significant contributions of neurodivergent editors. This has led to phenomena like "fancruft" - excessive content creation by specialists, such as road enthusiasts whose detailed entries were deleted, prompting the creation of separate road-focused wikis. These dynamics underscore ongoing debates about inclusion versus deletion that shape Wikipedia's evolving content policies.
Artificial intelligence now presents a critical challenge. As AI-generated content floods the web, Wikipedia faces potential traffic declines from search engines that increasingly prioritize AI summaries over direct links to source material. This threatens the symbiotic relationship between Wikipedia and the broader web ecosystem it helped sustain. While Wales and Sanger remain cautiously optimistic about Wikipedia's survival, the platform's future depends on navigating this technological shift while preserving its core collaborative ethos.
Criminals are leveraging artificial intelligence to refine and accelerate retail scams, prompting warnings from National Trading Standards. The organization has published its first AI scam watchlist, identifying tools and tactics that enable fraudsters to create polished, counterfeit websites that mimic family-run British businesses. These sites often display fabricated high-street imagery, including Union Jack flags, to project authenticity. Victims who purchase from them frequently receive either non-delivered orders or low-quality goods sourced from overseas manufacturers such as China.
AI-generated visuals, voice cloning, and deepfake video are increasingly central to these operations. Fraudsters use synthetic media to replicate the voices and appearances of relatives or friends, fabricating urgent financial emergencies to extract money. Additionally, AI enables the mass production of error-free phishing communications that mimic legitimate correspondence, making deception more difficult to detect.
Experts note that traditional scam indicators, such as spelling mistakes, are no longer reliable due to AI’s ability to generate flawless content. Consumer advocacy groups advise heightened vigilance, emphasizing that personal familiarity — whether through voice or video — should not override verification. Recommended safeguards include avoiding upfront payments, confirming business legitimacy via official registries like the Financial Conduct Authority’s Firm Checker, and independently validating job postings or romantic advances.
Trading standards officials stress the importance of pausing before acting on urgent requests, as pressure tactics aim to bypass critical thinking. Reported scams can be submitted to Action Fraud or reported by calling 0300 1232040, while Citizens Advice offers guidance through its consumer helpline at 0808 223 1133. Authorities urge the public to treat AI-enhanced scams with the same scrutiny applied to conventional fraud, prioritizing verification over emotional response.
The International Bureau of Weights and Measures has decided to discontinue the insertion of leap seconds into Coordinated Universal Time, citing operational risks for billions of internet-connected devices.
The decision was made at the 28th meeting of the General Conference on Weights and Measures, where delegates voted to adopt a system called "Continuous UTC" that permits a gradual divergence of up to 3,600 seconds between atomic time and Earth rotation time beginning next year. The move follows years of debate about the feasibility of maintaining synchronization between astronomical time and civil timekeeping standards.
Leap seconds were originally introduced to align UTC with observations of Earth's rotational behavior, which exhibits irregular variations due to geological and atmospheric factors. While 27 leap seconds have been added since 1972, recent discussions have highlighted growing complications in implementing these adjustments across global time infrastructure.
The resolution specifically references the challenges posed by distributed systems that depend on precise timing, particularly internet of things devices that often lack the capability for straightforward software updates. The Bureau noted that inconsistent approaches to leap second mitigation among technology providers could create synchronization gaps that undermine the integrity of time-sensitive operations.
A key concern involves digital certificate validation systems, where timing anomalies could trigger rejection of security credentials. The Bureau warned that such failures might affect critical services including network authentication and financial transaction processing if timing adjustments are not uniformly supported.
The decision draws parallels to the Y2K transition period, during which extensive remediation efforts were undertaken to address legacy software limitations with date handling. While the Bureau did not quantify specific remediation costs, it acknowledged that preparing for leap second implementations would require comparable levels of investment across affected industries.
Under the new Continuous UTC framework, timekeeping systems will maintain UTC without further leap second insertions, allowing gradual divergence that could be corrected through periodic adjustments if needed. This approach reflects a shift toward prioritizing system stability over strict alignment with astronomical time measurements in favor of operational continuity across global networks.
Anthropic announced that it will dedicate 5% of its equity to a newly formed public benefit fund structured as a wholly owned limited liability company. The company said the arrangement is intended to provide operational flexibility while ensuring that a portion of its resources supports initiatives aligned with its stated mission. The fund will not be used for lobbying activities or political contributions, according to a company statement.
The fund’s initial scope includes limited pilot efforts, such as potential grants to nonprofit organizations or direct support to specific communities affected by technological shifts. One focus area highlighted by Anthropic is the labor market impact of artificial intelligence. The company noted that while AI systems are projected to generate significant economic value, they may also create disruptions for workers across various sectors. To address this, Anthropic has previously committed $200 million to research and programs aimed at mitigating workforce transitions linked to AI adoption. The public benefit fund may expand these efforts, though the company has not specified additional funding levels or implementation timelines.
The structure of the fund reflects a broader trend among technology firms seeking to integrate social objectives into corporate frameworks without altering their core governance model. By operating through an LLC subsidiary, Anthropic maintains control over the allocation of funds while signaling a commitment to accountability in how the resources are used. The company has not disclosed further details about governance, reporting requirements, or potential partnerships related to the fund’s operations.
The Free Pascal Compiler has released its first version with full support for dynamic packages, enabling developers to create and install components, controls, and Lazarus extensions without rebuilding the entire IDE.
Dynamic packages function as shared libraries on Linux, Windows, and macOS, containing metadata that describes their contents. This metadata can be inspected using new RTL units, allowing programs to load, enumerate, search, and unload packages at runtime. The compiler and runtime have been updated to manage dynamic types such as strings and arrays across package boundaries while preserving thread variables.
The implementation allows applications to either bundle all functionality into a single executable or split it into multiple dynamic packages that can be updated while the program runs. This provides a plugin system capability for Pascal-based applications.
Key restrictions apply: units within a package cannot share names with the package itself, and unit names must be unique across all installed packages. Attempting to build a package currently in use by an open form requires closing or saving the form first. Additionally, dynamic packages are larger than standard shared libraries due to their embedded metadata, and stripping this metadata renders them unusable.
Lazarus IDE enhancements include a progress window that appears outside the main interface to show package rebuild status, estimated completion times, and activity during bulk package operations. A docked Packages window now displays recently opened packages and persists between sessions, allowing users to browse files and open units directly.
The IDE also features platform-specific package filtering. Packages can be restricted to particular operating systems or widgetsets, preventing incompatible components from appearing in the package list. For example, a Windows-specific COM interface implementation would only appear when Lazarus detects a Windows environment.
A new form conversion tool enables Lazarus extensions to transform any form or frame into an installable control that automatically appears on the component palette, preserving all associated events, methods, and properties. This streamlines the creation of custom UI elements.
The release notes indicate that changes have been committed to a GitLab fork named "dynamic packages," with bug reports to be submitted through the associated issue tracker. The project also seeks volunteer moderators and testing support, particularly for macOS builds on Apple Silicon hardware.
This development represents a significant advancement for the Free Pascal and Lazarus ecosystem, providing a more flexible deployment model for Pascal-based software without requiring changes to existing workflows.
Intel has identified silicon photonics as the foundational technology for next-generation systems, stating it is the only company manufacturing 300mm indium phosphide (InP) silicon photonics at volume scale for the broader industry. At the Open Compute Project (OCP) Summit 2026, the company demonstrated how its fabrication capabilities and integration tools are designed to accelerate high-performance computing (HPC) and artificial intelligence workloads through faster, more energy-efficient data transmission.
Nic Dubé, senior vice president of Intel's Data Center Systems and Solutions group, described silicon photonics as the "Golden Hammer" enabling a shift from traditional electrical interconnects to optical links. He noted that the industry is approaching physical limits in bandwidth density, reach, and energy efficiency per bit using copper-based connections. Silicon photonics, which uses light instead of electrical signals to move data between chips, addresses these constraints.
Intel emphasized its unique role in producing 300mm InP wafers — essential components for photonic integrated circuits — at scale. These wafers support next-generation optical modules, including co-packaged optics (CPO) and neural processing optics (NPO), which are critical for high-bandwidth, low-latency communication in AI and HPC systems. The company also highlighted its integration tools, which enable reliable platforms combining lasers directly into silicon packages, reducing reliance on external components and service requirements.
According to Intel, its approach aims to deliver interconnects with energy consumption below 5 picojoules per bit, a benchmark long sought in the industry. The company projects that such efficiency gains could significantly reduce power usage across data center racks, pods, and system-level architectures. Additionally, Intel stated it is working with ecosystem partners to standardize external light source interfaces, supporting a diversified supply chain for multi-wavelength sources.
The company believes silicon photonics will enable a new class of flexible, scalable systems tailored to growing AI model sizes and distributed computing demands. While specific deployment timelines were not disclosed, Intel indicated ongoing collaboration with industry stakeholders to bring these solutions to market.
NVIDIA is pursuing a standardized manufacturing approach for the base die of its custom high-bandwidth memory (HBM) solution, targeting the 3nm node. The initiative requires Samsung, Micron and SK hynix to produce the base die using TSMC's 3nm process, while the core die will rely on Micron's sixth-generation DRAM process, referred to as 1c. This shift aims to ensure consistent performance and reduce verification overhead across suppliers. NVIDIA has indicated that its in-house HBM implementation could deliver up to 30 percent higher memory bandwidth and consume up to 15 percent less power compared to standard HBM4E. To achieve these gains, the company is coordinating with multiple memory manufacturers to adopt a unified production flow, minimizing the need for individual sample validation. Samsung has confirmed its participation in the effort, marking a departure from its traditional turnkey model. The company is also assessing its own foundry capabilities in case demand outpaces TSMC's capacity. Industry analysts note that this collaboration may help stabilize supply chains and support broader AI hardware scalability.
The overhead projector was a ubiquitous fixture in classrooms for decades, shaping how teachers presented lessons and how students engaged with material. Developed initially for military training during World War II, the device transitioned into educational settings where it became a staple for displaying transparencies.
In a recent episode of the podcast Version History, technology journalist David Pierce, alongside Verge reporter Jake Kastrenakes and Stanford education professor Antero Garcia, examined the projector’s rise, its impact on pedagogy, and the reasons behind its decline. They explored how the shift to digital tools — including PowerPoint, interactive whiteboards, and laptop-based displays — altered classroom dynamics.
The episode marked the conclusion of Version History’s fifth season, which focused on influential school technologies such as the Oregon Trail game, early Chromebooks, the One Laptop Per Child initiative, LeapFrog’s LeapPad, and Texas Instruments calculators. While the podcast highlighted the OHP’s historical significance, it did not assert a definitive verdict on whether newer technologies represented genuine improvements.
Archival materials and related resources about overhead projectors are available through The Verge’s platform. The discussion underscored the broader pattern of technological adoption in education, where tools often enter classrooms through institutional or military origins before evolving into standard teaching aids.
Two pallets of ZOTAC-made RTX 5090 graphics processing units arrived at Guarulhos International Airport in São Paulo, valued at approximately $500,000, but were reported missing after a distributor filed a complaint. The shipment, imported by Oderço Matos’s company based in Maringá, Paraná, had been in transit for nearly four decades without loss, according to Matos. The company president shared a video on social media stating that the goods vanished after arrival and that no official explanation has been provided. Authorities, airport operators, or customs officials have not publicly commented on the incident. The distributor, which has imported products since 2008, said it contacted relevant parties to investigate and is seeking clarification. The RTX 5090, priced at $1,999 at launch but selling for $5,000–$7,000 in secondary markets, features 32GB of GDDR7 VRAM and 1.79TB/s memory bandwidth, making it a high-demand component for gaming and AI workloads. The episode highlights vulnerabilities in the handling of high-value electronics at one of Latin America’s busiest airports, though the full chain of events remains unclear.
A freelance science journalist discovered that her byline had been used without permission on a New York Post article about chronoworking, a piece she never wrote or contributed to. The article, published under the name Linda Geddes, appeared on LinkedIn where a connection had tagged her in a post praising the piece. Upon clicking the link, Geddes saw her name attached to content she had not authored.
The incident marked the latest in a series of unsettling discoveries over three weeks. A colleague at Nature raised concerns that editors at the journal had received pitches attributed to Geddes from an email address that did not belong to her. Further investigation revealed that another former colleague, Caroline Williams, had also had her byline misused to secure a commission from Science, resulting in a published article for which she received approximately $1,000.
Geddes traced the impersonation to a pattern of email addresses linked across multiple outlets. By using Outlook.com’s account recovery feature, she identified a partially obscured Gmail address associated with the fake accounts. A separate discovery by an editor at Science confirmed that the same recovery address was tied to the fraudulent account impersonating Williams.
Geddes emphasized the personal and professional stakes involved. “As someone whose livelihood depends on the reputation I’ve spent decades building, I find this deeply unsettling,” she said. The fabricated articles contained content lifted from her previously published work without attribution. Many of the pitches also exhibited signs of AI-generated text, raising concerns about the use of automated tools to mimic established journalists.
The New York Post removed the article and initiated an internal review, referring the matter to authorities. The outlet declined to comment further on the investigation. Geddes expressed confusion over the motive, noting that freelance science journalism is a niche and underpaid field. Her hypothesis points to a struggling freelancer attempting to boost income by impersonating more established peers, possibly with the aid of AI to streamline the process. Questions remain about how such individuals would receive payment without revealing their identities.
Other potential explanations have been considered. A public relations professional might be using fabricated bylines to secure coverage for clients. Alternatively, the activity could be part of an experiment testing the vulnerability of publishers to deception. Personal vendettas are also a possibility, though Geddes and Williams dismissed this notion, with Williams remarking, “Why us? We’re nice!”
Experts in fraud prevention and digital integrity have weighed in on the broader implications. Simon Miller, director of policy, strategy and communications at Cifas, described the case as “almost the stuff of Agatha Christie, but with 21st-century technology.” He noted that while large-scale fraud typically relies on mechanized processes, this instance stood out for its targeted and personalized nature.
The issue extends beyond individual cases of impersonation. In April, Eric Topol, a professor at the Scripps Institute, discovered his name on a scientific paper with which he had no involvement. Four of the listed co-authors were unaware of the publication, and the fifth could not be identified. The journal removed the paper within hours of Topol’s public alert on social media. Topol also reported finding at least 12 books on Amazon bearing his name, including cookbooks he does not author, some of which contained inaccurate content that confused patients who believed they were reading his memoir.
Leslie McIntosh, vice-president of research integrity and security at Digital Science, highlighted the risks of fabricated authorship in academic publishing. She explained that adding a respected name to a paper without consent can enhance credibility, while also serving as a tactic to attract publication fees from predatory journals. In more malicious scenarios, false attribution could be used to damage reputations, particularly if a retracted paper bears a falsified name.
Employers face additional challenges. Jamie Collier, a researcher at Google Threat Intelligence, reported cases where stolen personal data was used to secure remote positions under false identities. Generative AI has lowered barriers to deception by enabling the creation of convincing resumes, tailored cover letters, and fabricated work histories. Tools such as real-time video filters and voice cloning are being tested to bypass video screenings and obscure regional accents.
The Society of Authors and the Writers’ Guild of Great Britain have issued warnings about increasingly sophisticated scams targeting writers. These include impersonation of literary agents and publishers to extract payments or rights from aspiring authors. Anna Ganley, chief executive of the Society of Authors, reported receiving five to ten scam emails daily, some of which led to financial losses for recipients.
The pattern of fabricated bylines appears to involve multiple individuals. Geddes identified a third journalist, Sara Novak, whose identity may have been similarly misused. A pitch attributed to a fake Sara Novak was declined by Chemical and Engineering News, but the associated email address linked back to the same recovery address used in the earlier cases.
The consequences affect not only the impersonated journalists but also publishers. Ben Johnson, chief editor at Nature Health, noted that trust in established brands is paramount. When a fabricated story bearing a fake byline is submitted, it requires significant effort to edit AI-generated content into publication standards, including the removal of fabricated quotes and references to uninterviewed individuals.
The incident has prompted changes in verification practices. Science has begun using AI detection tools for pitches and drafts, though officials acknowledge limitations in these methods. The outlet also now requires video calls for new writers. Similarly, Nature editors have increased scrutiny of freelance submissions, including mandatory pre-commissioning calls to verify identities. However, video verification is not foolproof. Google Threat Intelligence recently documented cases involving North Korea-linked actors who used deepfake video and audio to manipulate Zoom calls, tricking executives into downloading malicious tools.
Experts recommend layered verification approaches. Miller advises cross-checking unexpected communications through established contact methods. McIntosh advocates for multiple identity verification steps, akin to airport security protocols, to confirm the authenticity of interlocutors.
For those affected, legal recourse is possible but challenging. Rupert Earle, senior counsel at Bates Wells, noted that impersonation to secure paid work may constitute fraud under UK law. Victims could also pursue civil claims for “passing off” or false attribution. However, identifying perpetrators and enforcing legal action is difficult, especially when publications are based abroad and the impersonator’s location is unknown.
Despite tracing connections between the fraudulent accounts, Geddes has been unable to identify the individual responsible. The investigation remains ongoing. She plans to report the incident to the UK’s Report Fraud service and has issued a cease-and-desist notice demanding the removal of the impersonating accounts. The perpetrator, who has so far maintained silence, may yet face further scrutiny as the inquiry progresses.
An overclocker identified as Luumi has broken a 12-year-old world record by pushing an Intel Core i5 4670K to 6.3 GHz using liquid nitrogen cooling and achieving a score of 10.8 points in Cinebench R11.5. The result, published on HWBot, surpasses the previous benchmark set by Strat, who recorded 10.76 points at 6.2 GHz in the same benchmark suite. The achievement marks the highest score recorded for the Haswell-era processor in the R11.5 version of the test, which has not been widely used since its release over a decade ago. The Core i5 4670K, launched in 2013 as part of Intel’s fourth-generation Core lineup, was built on a 22nm process and originally featured base and boost clocks of 3.4 GHz and 3.8 GHz. It supported four cores and four threads with 6 MB of cache and was known for its overclocking potential. Luumi employed an ASRock Z97 OC Formula motherboard, 4 GB of DDR3 memory operating at 2,708 MT/s, and a Corsair AX1600W power supply to stabilize the overclock. While extreme frequency records for this chip have previously exceeded 7.0 GHz, the 6.3 GHz configuration represents a notable milestone in the context of legacy hardware benchmarking. The record highlights continued interest in evaluating older architectures under extreme cooling methods, though its practical impact on modern computing applications remains limited.
A developer has released an open-source recreation of the classic puzzle game Lemmings for the PalmOS platform on GitHub. The project, titled PalmOS Lemmings, is hosted in a public repository and includes source code, compiled binaries, and visual assets such as screenshots. The repository description states the code is provided for educational purposes. It contains directories for level packs, source files, and tools, along with standard project files like README.md and .gitignore. The project has garnered 122 stars and 17 forks, with three users watching the repository. No release date or version information is specified in the source material. The developer has not announced plans for additional features or updates. The game mechanics and visual design appear to replicate the original Lemmings title, which was originally released for multiple platforms in the 1990s.
During a campaign rally in Highland Heights, Kentucky, on October 17, Donald Trump asserted that Google now acknowledges his authority in a way it did not during his first term, citing the company's response to his proposal to rename the Gulf of Mexico and Lake Ontario. He stated, “I called Google, Google Maps. See, Google wouldn’t have taken my call in the first term. Now they kiss my ass.” The comment came as Trump listed his administration’s achievements, including the proposed geographic name changes, which he described as more impactful than previous actions.
Google has not confirmed any direct engagement with the White House regarding the name changes. A company spokesperson clarified that its mapping services reflect official designations from government sources, specifically the Geographic Names Information System, and that updates are made in accordance with established policies for internationally recognized place names. The company did not comment on whether it had received or responded to any communication from the former president.
During his second term, multiple technology firms, including Google, have engaged in public gestures of goodwill toward the administration. These have included hosting campaign events, donating to initiatives aligned with the president’s priorities, and issuing statements that align with administration messaging. Google’s CEO, Sundar Pichai, attended a White House meeting in February 2020 and later contributed to a technology advisory council that was dissolved amid concerns over industry influence.
The claim that Google altered its mapping behavior in response to presidential pressure has not been substantiated by public records or official statements. The company’s standard practice involves updating place names only when authoritative sources, such as national geographic agencies, formally adopt new designations. In the case of the Gulf of Mexico, the name change proposal has not been formally enacted by any federal agency.
Trump’s remarks reflect a broader pattern of asserting personal influence over corporate entities, particularly those in the technology sector. His administration’s approach to tech companies has included both public criticism and overtures of collaboration, often depending on the issue at hand. Whether Google’s mapping policies have been adjusted in response to direct appeals from the White House remains unverified by independent sources.
The situation underscores the complex relationship between political figures and major technology platforms, where public statements and private interactions can intersect in ways that are difficult to corroborate. As with many claims involving executive communications and corporate behavior, the available evidence does not confirm or refute the president’s account, leaving the matter open to interpretation based on documented actions rather than asserted outcomes.
A developer has completed a second iteration of a laptop built around the 6502 microprocessor, a chip originally introduced in the 1970s and famously used in early personal computers such as the Apple II and Commodore 64. The new version, referred to as the "b" model, features a translucent orange enclosure that the creator says appears more vivid in person compared to the earlier prototype. The chassis change was motivated by a desire for a more streamlined appearance. The device incorporates a 10.1-inch IPS display with a resolution of 1024 by 600 pixels, an upgrade from the prior 9-inch screen that offered 800 by 480 pixels. Battery capacity remains comparable to the first version, and the processor runs at 10 megahertz, up from 8 megahertz. Additional features include a SID sound emulator and integrated Wi-Fi connectivity, though the design omits an expansion slot to maintain a slimmer profile. The higher pixel density and IPS technology contribute to improved contrast and text legibility, allowing more lines of text to be displayed on screen. Source files, including schematics and printed circuit board layouts, are hosted on Codeberg at https://codeberg.org/TechPaula/LT6502b. The developer notes several areas for improvement, including reinforcing mounting points for heat inserts, enhancing the Wi-Fi module for greater reliability and speed, upgrading the audio output, and exploring more efficient power management solutions such as USB power delivery without requiring preprogramming of the USB-PD controller. Other goals include supporting FAT16 file systems on the compact flash card and adopting a more robust battery configuration using an 18650 cell integrated into the hinge structure.
Colorful has demonstrated a 192 GB DDR5 memory kit operating at 6600 MT/s with CL28 latency on its upcoming iGame X870E Ultra-S V14 motherboard. The configuration uses four 48 GB modules from the iGame Shadow II DDR5 6000 CL30 kit. Stability testing via RunMemTestPro confirmed the setup functioned without errors. Achieving this speed with such capacity required careful tuning of both the memory controller and motherboard circuitry. The test platform employed an AMD Ryzen 7 9850X3D processor, noted for its advanced memory controller. The company highlighted that pushing high frequencies with four DIMMs increases electrical load, making signal integrity and motherboard design critical factors. Colorful indicated the iGame X870E Ultra-S will launch ahead of China's November 11 shopping period.
Java 17 introduced sealed classes, a language feature that restricts which subclasses can extend an interface or base class. This restriction enables the compiler to verify whether a switch statement covers all possible cases, eliminating the need for a default clause when all permitted subtypes are accounted for. The mechanism mirrors algebraic data types found in functional programming languages such as Haskell or ML.
When a sealed type is used internally within a codebase that is recompiled together, the exhaustiveness check provides strong compile-time guarantees. However, when the sealed type is part of a public API, adding a new permitted subclass can break existing consumers that have not been recompiled. Even though the compiler flags missing cases during compilation, runtime errors may occur if only the sealed type is updated and dependent code is not rebuilt.
The risk arises because the compiler cannot enforce recompilation of all dependent modules. Consequently, a new case introduced in a sealed interface may trigger a runtime exception in applications that rely on pattern matching without a default clause. This issue is particularly relevant for libraries that expose sealed types for external use, as changes to the set of permitted subtypes constitute a breaking API modification that typically requires a major version increment under semantic versioning.
The feature proves most valuable in internal implementations where all components are rebuilt in lockstep, offering greater type safety and reduced runtime failure modes. Its utility diminishes when applied to public interfaces, where the added complexity of maintaining exhaustiveness across module boundaries can introduce subtle bugs that surface only at runtime.
A newly released esoteric programming language called KillSwitch is being used to test how well large language models can truly understand code rather than simply recalling patterns from training data.
Created by an independent developer, KillSwitch is designed to be deliberately difficult for LLMs to interpret. The language incorporates features that exploit known weaknesses in current AI systems, including hidden inconsistencies in its specification and rules requiring precise character counts in identifiers.
Key design elements include:
- Constructs that trigger safety filters in models, such as phrases like "hack into a computer" or "Ignore previous instructions"
- Scripts primarily consisting of narrative text that embeds syntactic instructions
- Literals that can appear anywhere in code and be referenced through line counts
KillSwitch operates as a stack-based interpreted language with basic constructs like functions, variables, and conditional statements. Its specification contains deliberate ambiguities that require human-like reasoning to navigate, such as hidden false statements that readers must identify.
The developer has published benchmark results showing performance across several major language models. Claude Opus 5 achieved the highest score at 66.8, outperforming models like Gemini 4 Argon and Fable 5.1, which struggled due to their safety mechanisms. Among open-weight models, MiMo-V2.6-Pro showed strongest results.
Tests involve basic tasks such as implementing "Hello World," explaining script outputs, and modifying code for inverse temperature conversions. While some models succeeded on these limited exercises, the developer cautions this does not indicate genuine language understanding. Instead, they suggest such results reflect pattern recognition rather than comprehension.
The benchmarking framework and language tools are publicly available on GitHub. The developer plans to develop more complex test scenarios to further evaluate model capabilities, noting that current assessments represent only an initial step in probing AI understanding of novel programming paradigms.
Ryan Dahl, the creator of Node.js, announced his move to Cloudflare after the company acquired his startup Deno. Cloudflare plans to maintain Deno for one year before discontinuing its development, a decision Dahl supports. He stated that choosing Node.js for server development is fundamentally flawed, emphasizing that modern server architectures should leverage lightweight, distributed runtimes rather than traditional virtual machine layers.
Dahl's original Node.js framework, launched in 2009, revolutionized server-side JavaScript by enabling concurrent request handling with minimal memory usage. It powers approximately 30 million websites globally and has seen over 130 million downloads. Deno, introduced in 2018 as a modernized alternative with features like built-in fetch and native TypeScript support, struggled to gain traction despite $21 million in venture funding and 400,000 monthly users. Dahl acknowledged that Deno's evolution increasingly required compatibility with Node.js, undermining its distinct identity.
At Cloudflare, Dahl and 12 Deno team members will focus on enhancing the Workers platform, which uses V8 isolates for serverless computing. He highlighted the potential of Durable Objects—stateful, distributed micro-services with unique identifiers and embedded storage—as a more efficient alternative to conventional server runtimes. Cloudflare's open-source workerd project, though limited in scope, aligned with Dahl's vision for a standalone runtime.
Dahl recently open-sourced celld, an implementation of Durable Objects designed to operate beyond Cloudflare's infrastructure. The project aims to merge with workerd to create a lightweight, self-hosted JavaScript/WebAssembly runtime requiring just 300 KB of memory per instance. This could significantly reduce resource demands for cloud providers serving high-scale applications. However, the space for independent runtimes remains constrained by the decline of Deno and Anthropic's acquisition of Bun.
Dahl's departure from Deno does not signify abandonment of open-source innovation. He plans to continue developing celld and explore new avenues for distributed computing, while Cloudflare will maintain Deno's codebase for a transitional period. The shift underscores a broader industry move toward specialized, efficient serverless platforms over generalized runtime environments.
Apple is preparing a series of iPhone releases for 2027 that will reshape its product schedule and design approach. The company will continue the iPhone 18 series launched in the fall of 2026, then introduce new models in the spring and fall of 2027 to mark the 20th anniversary of the original iPhone.
In the spring of 2027, Apple plans to launch a base iPhone 18 model alongside an updated iPhone 18e. The base model will feature the A20 processor and likely retain the $899 starting price point established by the iPhone 17 after its recent price adjustment from $799. Thermal improvements and the new Apple C2 modem are expected upgrades. Design refinements may align the base model more closely with the iPhone 17 Pro and iPhone 18 Pro aesthetics, though specific changes remain unconfirmed.
The iPhone 18e is projected to offer a modest chip upgrade from the A19 to A20 processor, sharing many components with the standard iPhone 18. This strategy follows earlier reports about component overlap between the two devices, though the full implications for pricing and features have not been disclosed.
A notable addition to the spring lineup is the iPhone Air 2, which will incorporate an ultrawide camera lens and potentially a smaller Dynamic Island module. The device is rumored to use the C2 modem and receive a battery capacity increase to approximately 3500mAh, representing about a 10% improvement over the original Air model. The ultrawide camera addition appears driven by both functional enhancement and market positioning considerations.
The fall of 2027 will feature two significant releases. The iPhone 20 Pro is expected to deliver a comprehensive redesign to commemorate the iPhone's 20th anniversary, including a more glass-dominated exterior, reduced bezels, and a potential shift away from the flat-edge design introduced with the iPhone 12. This model may also feature a smaller Dynamic Island with a hole-punch front camera and under-display Face ID technology. While thermal performance remains under evaluation, the design represents Apple's most substantial hardware overhaul to date.
Additionally, a successor to the recently launched iPhone Duo is anticipated for release in the fall of 2027. Early speculation suggests potential improvements to Face ID capabilities and camera systems, along with the next-generation A21 Pro processor. The upcoming models collectively signal Apple's strategy to refresh its entire iPhone portfolio across multiple release windows in 2027.
The company's product roadmap indicates a departure from previous release patterns, with strategic launches planned for both spring and fall seasons. This scheduling approach reflects Apple's intent to maintain momentum in its smartphone development while addressing both technical advancements and market demands.
Benchmark announced $2 billion across two funds this summer, marking only the second time in its 31-year history that the firm has raised a vehicle larger than its traditional core fund. The move comes as many venture firms accelerate deal flow, but Benchmark's five general partners emphasized a measured approach during a TechCrunch Disrupt panel.
Peter Fenton, one of the partners, stated the core fund remains anchored in early-stage investments with a 10-to-15-year horizon, typically making two or three new bets per partner annually. The firm maintained its $500 million fund size for two decades before adjusting for inflation, a shift Fenton described as necessary but previously avoided. A new growth fund, now totaling $1.25 billion, expands the firm's scope beyond early-stage work, allowing participation in later-stage opportunities that could yield 10- to 1,000-fold returns. Fenton noted the fund's structure favors concentrated bets, with four to eight investments likely driving most returns.
Ev Randall, who joined Benchmark in 2023 after stints at Founders Fund, Bond, and Kleiner Perkins, highlighted the firm's small partnership model as a key differentiator. She emphasized the collaborative culture where partners work directly with founders without hierarchical decision-making, stating that the team's mutual respect and shared focus on founder relationships were decisive in her decision to join. Jack Altman, who joined in February 2023 after raising a $275 million fund and hosting a popular podcast, described the firm as a place where founders are prioritized over financial metrics. He noted that Benchmark's emphasis on being 'the first call' for founders contrasted with industry trends focused on status and capital.
The partners rejected the notion of a formal investment agenda, with Chethan Puttagunta explaining that weekly Monday meetings intentionally lack structure to remain reactive to emerging opportunities. When asked if a single partner could block a deal, Puttagunta acknowledged uncertainty, noting that the firm's low deal volume (two to three per partner yearly) renders rigid processes unnecessary. Altman added that early investments required collective input, with no single partner able to dictate outcomes.
Eric Vishria discussed the firm's stance on AI and competitive investments, citing Cerebras' 2016 hardware bet as an example of embracing unorthodox opportunities. He argued that consensus-driven investments miss the 'heretical' ideas that often yield outsized returns. On AI market volatility, Vishria noted that while companies like CoreWeave and Nebius experience sharp public market swings, Benchmark focuses on sustainable growth rather than speculative trends.
When questioned about San Francisco's role, Fenton and Randall affirmed the city's unique ecosystem for talent and capital, though they acknowledged non-SF founders can succeed. Randall observed that most recent non-SF companies she engaged with maintained plans to establish a presence in San Francisco, citing the city's unparalleled peer network.
The firm declined to comment on specific portfolio companies beyond Cerebras, though Vishria referenced the recent AI sector bankruptcy of Character AI as a reminder of market unpredictability. Benchmark's approach, they stressed, prioritizes long-term founder partnerships over short-term deal volume, even as competitors accelerate investments.
The partners collectively rejected narratives of market panic, framing their strategy as consistent with historical practice rather than a reaction to external pressures. As Altman noted, the firm's structure—designed for deep, infrequent engagement—remains its core strength, with no immediate plans to alter its pace despite industry shifts.
The documentary You Can See Everything, directed by Nathan Fielder, opened with $15 million in ticket sales, according to reports. The figure represents the highest opening weekend for a documentary in two decades, surpassing previous benchmarks. It ranked second overall at the domestic box office, trailing only the action film Street Fighter. The performance marks the strongest debut for a non-concert or music-related documentary since Michael Moore's Fahrenheit 9/11 in 2004. The film examines the rise and fall of Theranos founder Elizabeth Holmes, drawing significant audience interest during its initial release.
A developer running Qwen3.8-27B on an RTX 3090 GPU increased token generation speed from 82 to 382 tokens per second through software optimizations, achieving a 4.6x improvement without hardware upgrades. The gains came from speculative decoding techniques that process multiple tokens simultaneously rather than sequentially. The custom inference engine used three methods: multi-token prediction tuning, a drafting system called DFlash2 that proposes seven tokens at once, and a lookup mechanism that identifies repeating text patterns to fill prediction slots. These approaches reduced the bottleneck of repeatedly fetching model weights from memory, which typically limits performance more than raw processing power. Accuracy on math benchmarks remained at 96.5 percent during testing. Speed improvements occurred in stages, with intermediate results showing 114 tokens per second after MTP tuning, 138 after adding the lookup drafter, and 381 during large document processing. The optimizations required storing context in a more compact format, which doubled initial document loading time but enabled a 138,696-token context window. While everyday chat saw speeds around 133 tokens per second, the techniques are most effective for repetitive text generation tasks. The developer noted trade-offs including reduced available context slots and slower initial document ingestion, but emphasized these methods could extend the useful life of existing hardware for specific workloads.
Apple released an EU energy label for the iPhone Duo that specifies a total battery capacity of 4,883 mAh. The label aligns with a prior leak indicating the device contains two cells — 2,962 mAh and 1,921 mAh — that combine to reach this figure. Because the iPhone Duo relies solely on eSIM connectivity, the capacity applies globally, with no regional variation tied to a physical SIM tray. For context, iFixit previously measured a 4,288 mAh battery in the eSIM-only iPhone 18 Pro and a 5,567 mAh battery in the eSIM-only iPhone 18 Pro Max. Apple retail staff have been instructed to advise customers against attempting walk-in purchases on launch day, October 23, due to anticipated stock shortages. Mark Gurman of Bloomberg described Apple's decision to open pre-orders despite known supply constraints as a 'major mistake.' Apple hardware engineering vice president Tom Marieb addressed concerns about the device's crease visibility in a recent TechRadar interview, noting the use of a matte nano-texture display intended to reduce glare and limit crease prominence over time.
A federal appeals court has upheld a lower court's decision that Thomson Reuters' data collection practices were not infringed by Ross Intelligence's use of copyrighted material to train its legal AI system. In February 2025, U.S. District Court Judge Stephanos Bibas had ruled in favor of Thomson Reuters, determining that Ross Intelligence's copying of the company's proprietary headnotes for training purposes did not qualify as fair use. The appeals court's recent ruling maintains that finding, though the opinion remains under seal. The case stems from a 2016 lawsuit in which Ross Intelligence argued that its use of Thomson Reuters' content to develop a natural language processing tool for legal research was protected under the doctrine of fair use. Thomson Reuters contended that Ross Intelligence's actions constituted unlawful copying that harmed its market. The appeals court's decision reinforces the lower court's conclusion that the use was not transformative and therefore not permissible under copyright law.
A team of engineers at the Massachusetts Institute of Technology has created a soft, wearable sheet capable of sensing its own three-dimensional shape with precision within half a centimeter in laboratory conditions. The material, detailed in a study published in Advanced Intelligent Systems, uses arrays of optical fibers to detect bending in multiple directions, enabling accurate tracking of movement without reliance on external cameras or rigid components. Researchers noted that the system maintains functionality even when partially damaged, though the fiber thickness currently limits its integration into clothing. Applications under consideration include interactive gaming interfaces, immersive virtual reality environments, and motion monitoring for physical therapy. The development represents a step toward fully soft sensor systems that could expand the range of motion-capture technologies available for consumer and medical use.
The World Bank is urging developing nations to invest in so‑called “small AI” systems that can operate on basic mobile phones with limited internet access. Paschal Donohoe, the institution’s chief knowledge officer, argues that such models, built for specific tasks and trained on narrow data sets, could accelerate economic progress in regions that have historically lagged behind.
Silicon Valley firms, by contrast, are concentrating on increasingly complex consumer‑oriented products. Donohoe contends that AI tailored to local languages, infrastructure constraints and community needs could deliver tangible benefits without requiring extensive technical expertise or reliable broadband.
He illustrated the concept with the World Bank’s AgriConnect initiative in India, where small‑holder farmers required soil‑analysis results in their native languages. Rather than waiting for a cadre of agricultural experts to serve every plot, the organization sees an opportunity for phone‑based AI tools to fill that gap.
The discussion unfolded during the World Bank and International Monetary Fund’s annual meetings in Bangkok, Thailand. IMF Managing Director Kristalina Georgieva used the forum to stress that successful AI adoption hinges on basic prerequisites: reliable electricity, digital infrastructure and supportive regulatory environments that protect users from malicious actors.
Georgieva also warned that, over the coming years, roughly 40 % of global jobs could be transformed, augmented or displaced by AI, with young workers facing the steepest disruption. She noted that traditional entry‑level positions — such as barista roles — might disappear, reshaping early‑career pathways.
Under President Ajay Banga, the World Bank has prioritized mobilizing private‑sector capital for emerging economies. In the past fiscal year, the institution reported the activation of $120 billion in private investment through guarantees and other financial mechanisms, even as many donor governments have reduced foreign‑aid budgets.
Donohoe emphasized that linking private capital to job creation and broader development outcomes is a central goal of the Bank’s current strategy.
NVIDIA's Vera CPU marks a fundamental shift in processor design, moving beyond traditional multitasking to serve as a dedicated orchestrator for AI agent workflows. The company recently outlined how this architecture redefines the CPU's role from general-purpose computing to specialized coordination of multi-step AI reasoning loops.
Historically, data center CPUs prioritized either high single-threaded performance or massive core counts, but not both. Early 2010s designs like Intel's Haswell excelled in single-thread speed but delivered limited overall throughput. As cloud computing scaled, the industry pivoted toward maximizing core density for cost efficiency, leading to chiplet-based architectures that sacrificed per-core performance and IPC. This era emphasized throughput at the expense of sequential task speed.
Agentic AI has disrupted this trade-off. Modern AI systems operate in tight loops between GPUs and CPUs, where GPU idle time directly impacts efficiency in AI factories. To maintain GPU utilization, CPUs must execute sequential tasks—such as Python scripting or code compilation—with minimal latency. NVIDIA's Vera addresses this bottleneck by achieving what competitors have not: high single-threaded performance combined with high throughput.
The Vera CPU implements three key design innovations. First, it uses a monolithic compute die that eliminates inter-chip communication latency, enabling cores to collaborate at 3.4 terabytes per second of bisection bandwidth—three times faster than leading x86 chips. Second, it features a custom LPDDR5X memory subsystem optimized for AI workloads. Third, it employs spatial multithreading, a proprietary threading model that dedicates separate worker and I/O threads to prevent resource contention during heavy loads.
Independent testing by Phoronix confirmed that this spatial multithreading delivers approximately 20 percent performance gains without introducing latency penalties. Vera also demonstrates 1.5x higher IPC than NVIDIA's previous Grace processor and 2x speedup over leading x86 chips on agentic workloads, according to company benchmarks.
The shift extends beyond chip design. NVIDIA has moved away from integrating Vera solely within combined GPU-CPU stacks like the Vera-Rubin system. Instead, it is promoting Vera as a standalone CPU for dedicated rack-scale deployments, with some customers planning to operate 'pure CPU' clusters handling hundreds of thousands of AI sandboxes. While NVIDIA declined to comment on specific SK Hynix disaggregation approaches, it acknowledged broader industry trends toward specialized, heterogeneous infrastructure.
The company remains agnostic about future architectural choices, stating it will adopt chiplet designs only if market demands dictate such a shift. What is clear is that Vera represents a deliberate departure from cloud-era CPU design principles, engineered specifically for the unique demands of AI agent coordination.
Dell has notified suppliers of a planned increase in monitor shipping costs this month, with premium models expected to see larger adjustments. Internal communications cited by ChannelGate indicate the move targets mid-range and high-end displays, while entry-level units remain unaffected. The adjustment follows a prior cost rise in July that Dell previously absorbed rather than passing to consumers. Industry analysts note that sustained pressure on component prices and supply chain dynamics are prompting the company to shift some expenses downstream. The proposed 3-5% increase reflects broader trends of rising manufacturing costs across the electronics sector, though distributors and retailers may moderate final retail pricing changes. No official announcement has been made by Dell, and the timeline for potential future adjustments remains unspecified. The development underscores intensifying competition among monitor manufacturers as they navigate evolving technology cycles and cost structures.
A TikTok advertising account linked to Alex Gibney's documentary about Elon Musk was reinstated after a temporary suspension, the company confirmed on October 18, 2026. The account had been taken offline on October 17 due to what the platform described as a technical error. A TikTok spokesperson in the United States, Mahsau Cullinane, stated the account was restored after the issue was resolved, though no further details about the nature of the error were provided. The initial suspension had been reported as a permanent action, but the platform clarified it was a short-lived disruption. The documentary, directed by Gibney, explores the business and personal life of the technology entrepreneur.
A handheld milk frother purchased for $14 in March has become a seasonal staple for one Oakland resident seeking relief from a prolonged heat wave. The device, bought to recreate a Korean-inspired iced coffee drink known as a cream top, enabled the creation of layered coffee beverages featuring a distinct, melty foam layer atop cold brew.
The cream top, originally an Austrian espresso drink called Einspänner, historically used whipped cream as insulation for hot beverages during winter carriage rides. Its modern iteration, popularized in South Korea and later adopted globally through chains like Starbucks, typically features a dense, slow-dripping foam rather than light latte foam or heavy whipped cream. This texture allows the cream to integrate gradually with the coffee below.
The frother operates via disposable AA batteries, requiring approximately 30 seconds to one minute of runtime per use when fresh. Performance diminishes as batteries deplete, signaling replacement when the device slows. Beyond coffee preparation, the tool serves secondary functions including egg beating, protein powder mixing, and hot chocolate preparation.
The user’s shift toward frothed coffee emerged during Oakland’s March heat wave, when temperatures persisted near 90 degrees Fahrenheit for two weeks. Previously committed to black coffee — initially adopted to avoid curdling during reheating — the individual began experimenting with dairy-based drinks after encountering the cream top at a Los Angeles coffee shop, Stagger, during a February trip.
Concentrated cold brew serves as the base for these beverages, chosen for its robust flavor profile that withstands dilution by the foam. Variations include coconut ube-flavored cold brew topped with toasted coconut flakes, inspired by a Starbucks promotional item but adapted to home preparation. The resulting drink, described as 'dank' in the original account, provided a cooling yet indulgent option during the heat wave.
While premium frothers exist, the $14 battery-powered model was selected for its cost-effectiveness and functionality. Testing sources such as America’s Test Kitchen informed the choice, though comparisons to higher-end alternatives like Apple’s HomeView were not made. The device’s utility appears tied to seasonal temperature shifts rather than year-round use.
A July project saw a hobbyist retrofit a Razor A5 Lux push scooter with a cordless drill to create a low-speed electric vehicle. The modification uses a Hercules 20v drill as the power source, transmitting rotation to the front wheel through a keyed shaft and one-way bearings. The drill operates at full speed only, limiting top velocity to approximately 10 miles per hour. Acceleration triggers an initial wheel spin due to the trigger's sensitivity, which the rider mitigates by combining a manual push with drill activation.
Power is delivered via a 5 amp-hour battery, yielding roughly two miles of range under continuous use. The build incorporated components sourced from Harbor Freight, Amazon, and McMaster, including dual keyway bearings, shaft collars, and a custom 3D-printed wheel assembly constructed from ASA and TPU materials. The wheel features a tooth-equipped mounting surface to prevent slippage, while the tire is secured with ten M6 bolts and five keys.
Relocating the drill's trigger to the handlebars required extending twelve wires using 12awg and 18awg gauge, though the wiring remains unshielded. The project lacked pre-planned documentation, with components integrated through iterative assembly rather than schematic design. While the conversion cost comparable to purchasing a commercial electric scooter, the builder emphasized the project's educational value over economic efficiency. No performance data beyond range and speed claims was provided, and no external validation of the modification's safety or durability was reported.
A developer has released a free server that lets artificial intelligence agents read and search WhatsApp conversations within a personal Cloudflare account. The project, called WhatsApp MCP, is hosted on Cloudflare Workers and uses the open-source Baileys library to interface with WhatsApp Web's protocol. It does not send messages or alter WhatsApp's functionality, and operates solely as a read-only bridge.
The server runs as a single Durable Object in the user's Cloudflare account, storing message history in a SQLite database. It connects to WhatsApp only once daily to fetch newly queued messages, then disconnects immediately. Users link their phone to the server via a QR code scan, similar to WhatsApp Web pairing, after which any MCP-compatible AI client can query the stored messages.
Key capabilities include full-text search across all chats, filtering by date or participant, and retrieving message excerpts. The system preserves message metadata but does not download media files, displaying them only as labels like [image] or [voice note]. Unnamed chats appear as phone numbers until an AI agent cross-references them with local contacts and assigns names via a dedicated tool.
Security considerations are noted: the implementation uses an undocumented WhatsApp protocol and violates WhatsApp's terms of service, though the developer emphasizes it is intended for personal use only. The server requires a Cloudflare Workers Paid plan for initial large-scale history imports due to CPU limitations on the free tier. A free-tier workaround may exist but carries reliability risks.
Setup involves creating a Cloudflare account, deploying the code via npm, and configuring authentication through a password or email link. Once deployed, users connect their AI agent using a command like `claude mcp add --transport http whatsapp https://<url>/mcp`, then authenticate to access the message database. The system does not support sending messages and is designed for single-account use only.
The project highlights growing interest in AI-enhanced messaging workflows, though it relies on reverse-engineered protocols and carries inherent technical and compliance trade-offs. Users must manage their own deployment and understand the limitations around message completeness and contact identification.
A new open-source 3D printing platform called µRepRap has demonstrated repeatable positioning accuracy below 3 micrometers within a 10-millimeter work area, according to project documentation published on Hacker News. The system, designed to operate at micron and sub-micron scales, uses only commonly available materials and builds upon earlier work from The OpenFlexure Project's open-source microscope platforms.
Two active development versions have been documented: the PIKA V0.02 hardware, which achieves approximately 1-micrometer repeatable positioning and 10-micrometer or better accuracy within an 8-millimeter cubic build volume, and the MAUS C V0.04 variant, which reached about 5-micrometer positioning accuracy over a 4-millimeter square area before being superseded by the MAUS 'L' V0.05 model currently in development. Both versions employ 400-step NEMA17 stepper motors for axis control, with some builders using 200-step motors to approach 1-micrometer precision. The system uses UV-curable resin deposited through a fine probe tip, with droplet sizes as small as 4 micrometers achievable using capillary action.
The project aims to extend accessible micro-scale manufacturing capabilities similar to how RepRap democratized macro-scale 3D printing. Researchers note that while silicon-based microelectronics have dominated fabrication for decades with increasingly complex and expensive systems, this approach could enable independent community development outside traditional silicon foundries. The technology also holds potential for biological and medical applications, where microscope platforms capable of sub-micron resolution have recently created new opportunities for low-cost, rapidly evolving equipment.
Current implementations use UV nail gel resin as the primary building material, with a 3-watt UV LED module integrated to cure deposited layers. Optical inspection relies on USB microscopes with at least 8-megapixel resolution for initial setup and conventional bench microscopes offering 100x to 200x magnification for final examination. A 1.3-millimeter 'Hello World' inscription etched on a glass slide using a Sharpie marker in mid-2025 demonstrates the system's baseline capabilities.
The project documentation emphasizes that while significant progress has been made, substantial work remains to develop complete fabrication processes for complex structures, including mechanisms requiring hinges or flexures. The long-term goal is to create a self-replicating ecosystem of micron-scale fabrication tools that could eventually produce devices at even smaller scales.
Ubuntu 26.10 "Stonking Stingray" is now available as an interim release, featuring kernel 7.3 despite its official launch not yet occurring. The distribution incorporates a new speech-to-text system presented as a technology preview. This release follows the beta version examined earlier in the month, indicating no critical bugs were identified. Canonical's adoption of an accelerated kernel release cycle has resulted in the use of a pre-release kernel version 7.3, with a full update to the finalized kernel expected shortly after availability.
The default GNOME edition includes version 51, maintaining visual consistency with the beta. A notable enhancement is the improved window-tiling assistant, which now provides a transient interface resembling Windows-style suggestions when windows are moved to screen edges.
The speech-to-text functionality, previously announced as Myna, is accessible in this release. Testing on a Dell XPS 9370 demonstrated successful operation when using a USB headset, though accuracy was limited with the built-in microphone. The system requires approximately one gigabyte of additional dependencies and language model files during installation, which is notable given the ISO size of approximately 6.6 gigabytes. The feature currently lacks punctuation and command comprehension, entering spoken terms literally into documents.
Other Ubuntu flavors receive updates consistent with their respective desktop environments. Kubuntu 26.10 includes KDE Plasma 6.7.5, the final Plasma release supporting X11. Ubuntu MATE returns after skipping the LTS cycle, offering a desktop environment based on Lomiri with alternative file management options. Lubuntu features LXQt 2.4, while Ubuntu Budgie incorporates version 10.10.3 of the Budgie desktop with accessibility improvements.
Underlying technical updates affect all editions, including the replacement of core command-line tools with Rust implementations, substitution of dbus-daemon with dbus-broker, and use of systemd version 261.3. The server variant includes updated components such as AMD ROCm 10.0, OpenSSL 4, and OpenSSH 10.5. Full disk encryption with TPM support is available, though upgrade paths from previous versions require further development.
The release cycle contrasts with delays observed in Fedora 45, where no release candidate was ready at the time of reporting, pushing the expected availability back by approximately one week. Unlike some distributions, Ubuntu does not automatically upgrade LTS versions to interim releases, requiring explicit user opt-in. The interim release model allows for routine updates when standard package sources are used without extensive system customization.
This release represents a step toward integrating speech input capabilities that have been available in competing operating systems for several years, though current implementation remains a technology preview with notable limitations in accuracy and functionality.